Open ASICseer opened 7 years ago
Good idea. Will consider offering this option.
@mbevand
It may be a good idea to allow--instances auto
or something similar, to programmatically find the total vram per gpu and assign as many instances as can fit into it.
Example php code / hack way to do it (I am sure you will have a better way).
<?php
function return_threads(){
$vram_array = explode(" ",file_get_contents("vrams.file")); // vrams.file = "4 2 2 4"
$count = 0;
foreach($vram_array as $vram){
$gpus[] = $count;
$count++;
if($vram >= 3.5) {
$threads[] = 2;
} else {
$threads[] = 1;
}
}
$instances = implode(",",$threads);
$use_gpus = implode(",",$gpus);
return array($use_gpus,$instances);
}
?>
I've noticed a GPU may have enough memory for 3, 4, or more instances but may still run faster with 2.
Many people run mixed GPU rigs of 2gb cards and 4gb cards. Please add feature to define instances like:
--instances 1,2,2,1
This will allow users to define instances on a per-gpu basis. Right now, the only work around is to run separate miner instances with different flags.
Thank you!